| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
a Dep. of Crop Sci., North Carolina State Univ., Box 7620, Raleigh, NC 27695-7620
b Dep. of Crop Sci., North Carolina State Univ., Vernon James Res. and Ext. Cent., 207 Research Rd., Plymouth, NC 27962
* Corresponding author (mflowers{at}cropserv1.cropsci.ncsu.edu)
Received for publication May 19, 2000.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
r 0.87). The inclusion of within-field tiller density references resulted in a high correlation (r = 0.88) between relative tiller density and relative NIR digital counts across environments. Using relative NIR digital counts to predict tiller density would have resulted in the correct N recommendation 82% of the time.
Abbreviations: B, blue • DVI, difference vegetation index • G, green • GS, growth stage • NDVI, normalized difference vegetation index • NIR, near infrared • OSAVI, optimized soil-adjusted vegetation index • R, red • RVI, ratio vegetation index • SAVI, soil-adjusted vegetation index • TRS, Tidewater Research Station
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Scharf and Alley (1993) developed a N management strategy using these two critical periods for wheat. They first determined the whole-field tiller density at GS 25. If the tiller density was low (<1000 tillers m-2), applying N at GS 25 improved grain yields and economic returns. At GS 30, they found that a tissue test could be used to optimize further N applications. Weisz et al. (2001) examined the critical range for GS-25 N application for a wide range of tiller densities. They found that when tiller density was <540 tillers m-2, applying all of the spring N at GS 25 produced grain yields that were greater than those obtained by applying all of the N at GS 30. However, when tiller density was >540 tillers m-2, applying all of the spring N at GS 25 produced grain yields that were less than those obtained when all of the N was applied at GS 30. These studies show that to obtain maximum grain yield, wheat growers must know the tiller density at GS 25 and be able to respond quickly by applying N when needed.
In spite of the benefits of timing N based on GS-25 tiller density, few growers currently use this system. There are two reasons for this: Tillers are difficult and time consuming to count, and tiller density is highly variable across a field. Variable soil characteristics, crop residues, planter problems, and drainage patterns can all significantly impact tiller density. Therefore, to obtain an accurate average tiller density, frequent and intensive measurements must be made across all areas of the field. Furthermore, the time involved in such intensive sampling can delay the N application. Clearly, other methods for determining tiller density must be found if this information is to be used successfully to adjust spring N applications and increase grain yield.
One solution to this problem may be to use remote sensing in the form of aerial images or photographs to determine tiller densities across a wheat field. Aerial photographs have been used to examine crop management (Schuler et al., 1999). For example, the green color of plant canopies is associated with leaf chlorophyll content, which affects the amount of light absorbed or reflected in the visible wavelengths (400–700 nm) (Thomas and Gausman, 1977; Maas and Dunlap, 1989). Aerial photography has been used to detect several factors, including diseases (Colwell, 1956), insect damage (Wildman, 1982), and N deficiency (Blackmer et al., 1996). Several scientists have noted a relationship between biomass or leaf area and reflected radiation (Lukina et al., 1999). Wanjura and Hatfield (1987) did an extensive study of the relationship between crop biomass and spectral vegetation indices derived from reflected solar radiation. They found that there was a strong correlation between total biomass and the normalized difference vegetation index (NDVI) at high levels of canopy biomass and leaf area. However, at low levels of canopy biomass, the sum of near infrared (NIR) and red (R) wavelengths was a more accurate predictor of biomass. Wiegand et al. (1994) found that spectral reflectance in the NIR, R, and green (G) wavelengths could individually be used to estimate biomass, leaf area index, and crop yield. In wheat, Wood et al. (1999) reported correlations (0.57 r 0.95) between NDVI and GS-25 tiller density. A strong correlation (0.62 r 0.80) between GS-30 biomass and plant N spectral index, which is the inverse of NDVI, was reported by Stone et al. (1996).
Given the positive results of Wood et al. (1999) and Stone et al. (1996), our first objective was to determine if a spectral index or digital counts in the NIR, R, G, or blue (B) wavelengths derived from a color or false color infrared aerial photograph might be used to determine tiller density at GS 25 across environments to make N application decisions.
There are many factors that can influence the measured reflectance in the visible and NIR spectrums and could complicate the use of remotely sensed data to determine tiller density across environments. Weed populations (Menges et al., 1985), diseases (Colwell, 1956; Hatfield, 1990), insect damage and water stress (Wildman, 1982), varietal differences (Hatfield, 1990; Stone et al., 1996), and plant nutrition (Thomas and Oerther, 1972; Wildman, 1982) can all affect the measured reflectance. The soil surface can change the amplitude and wavelengths of reflected radiation (Lukina et al., 1999), and atmospheric conditions (Jackson et al., 1983) can influence the amount of solar radiation reaching the ground. Because of the large number of factors that could potentially influence the relationship between remotely sensed data and crop conditions, Blackmer et al. (1996) concluded that ground observations (truthing) or inclusion of known reference conditions could benefit the use of remote sensing across environments. Specifically, Blackmer at al. (1996) proposed using within-field references and a relationship between relative grain yield and digital counts of R, B, and G wavelengths relative to a within-field reference to detect N stress in corn. Consequently, our second objective was to determine if a procedure analogous to that described by Blackmer et al. (1996) would improve the potential use of remotely sensed data to predict tiller density and make N application decisions.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
At the TRS and Winslow farm sites, GS-25 tiller density was determined at two locations within each treatment strip by averaging two 1-m sections of row at each location. At the Whitehat farm site, tiller density was determined at five random locations along each of four transects by averaging two 1-m sections of row at each location. Latitude and longitude were determined for all sample locations and aerial targets placed at each field corner using a differential global positioning system (DGPS) receiver with 1-m accuracy (Trimble AG 132, Trimble Navigation, Sunnyvale, CA). On the same days that tiller densities were measured (13 Feb. 1998 for TRS, 5 Mar. 1998 for Whitehat, and 5 Feb. 1999 for both TRS and Winslow locations), aerial photographs were taken from a belly mounted platform using a 35-mm Canon 81 camera (Canon USA, Lake Success, NY) for the color infrared film and a 35-mm Nikon 6000 camera (Nikon, Melville, NY) for the color film. Aerial photographs were taken as low as possible while insuring each site was contained within a single photograph. This resulted in a range of heights (
854 m) across sites. All aerial photographs were taken on cloudless days between 1200 and 1400 h standard time. Kodak Ektachrome 153 film along with a Kodak Wratten gelatin filter no. 15 (Eastman Kodak Co., Rochester, NY) were used for the color infrared images and Kodak Ektachrome Elite 200 film (Eastman Kodak Co., Rochester, NY) was used for the color images.
Photographic analysis was done using a positive false color slide for the color infrared film and a color slide for the color images. Slides were digitized using the procedure described by Blackmer et al. (1996) with a Konica slide scanner (Konica Q-Scan, Konica Corp., Mahwah, NJ) and the software package Adobe Photoshop v. 4.0 (Adobe Syst., San Jose, CA). The size of the image was scanned with a resolution of 47 pixels mm-1, with each pixel representing a range of 0.07 to 0.41 m2 of ground area. The range in ground area was due to differences in altitude when the image was taken. Comparisons in all cases were limited to within a given photograph. The digitized images were orthorectified in ERDAS Imagine (ERDAS, 1997) using the latitude and longitude of the aerial targets.
Color infrared film emulsions respond to light within the visible and NIR (490–900 nm) regions of the electromagnetic spectrum while color film responds within the visible (400–700 nm) regions. The Konica scanner produced a RGB image with 24-bit true color and three bands (8 bit R, 8 bit G, and 8 bit B). At each location on the image, the primary color value represented RGB digital counts within the range of 0 to 255. These color values are directly proportional to the total light reflected from the scene. In the case of the color infrared film, Band 1 covered the NIR spectrum (>700 nm). The other two bands captured light from the R (Band 2, 550–700 nm) and G (Band 3, 490–550 nm) visible wavelengths. For the color film, Band 1 captured light from the R spectrum (590–730 nm), Band 2 captured light from the G spectrum (490–590 nm), and Band 3 captured light from the B spectrum (<490 nm).
Using ERDAS Imagine, the locations of each tiller density sample were overlaid on the orthorectified digitized photograph and buffered (expanded) to represent an area 1.07 m2. Digital counts for each color band were averaged across this area, and these average values were used to represent the spectral reflectance at that location (ERDAS, 1997). In addition to an examination of the digital counts for each band, a normalized NIR value (Jain, 1989) was derived such that:
 | (1) |
A NDVI (Yang and Anderson, 1999) was determined using the digital counts from the NIR and R bands from the infrared image such that:
 | (2) |
A ratio vegetation index (RVI; Jordan, 1969) and a difference vegetation index (DVI; Tucker, 1979) were also calculated as:
 | (3) |
 | (4) |
A soil-adjusted vegetation index (SAVI; Huete, 1988) and an optimized soil-adjusted vegetation index (OSAVI; Rondeaux et al., 1996) were calculated as:
 | (5) |
 | (6) |
Finally, the digital counts for the NIR and R bands were summed (Wanjura and Hatfield, 1987). The average digital counts in NIR, R, G, B, the normalized NIR, NDVI, RVI, DVI, SAVI, OSAVI, and NIR + R were compared with tiller density (as Pearson correlations) in SAS (SAS, 1998) to examine any relationships.
To determine if differences across environments could be removed using within-field reference values, a procedure analogous to that proposed by Blackmer et al. (1996) was used. This procedure consisted of two steps for each data set. First, the NIR digital count at the lowest tiller density (126, 86, 0, and 0 tillers m-2 for TRS in 1998, Whitehat, TRS in 1999, and Winslow, respectively) was determined. These NIR digital count values for each location and year were then subtracted from all NIR digital count values. Once each NIR digital count value was corrected for the low tiller influence, it was divided by the difference between the NIR digital count value measured at the plots with the highest tiller density and the NIR digital count value measured at low tiller density. This resulted in relative NIR digital count values ranging from approximately 0.0 to 1.0. Secondly, relative tiller density for each location was calculated by subtracting the low tiller density from all tiller density values and then dividing by the difference between the highest tiller density and the lowest tiller density at that location. Relative tiller numbers were compared with relative NIR digital counts in SAS (SAS, 1998) to determine if a single relationship existed across environments.
Finally, the accuracy of GS-25 N recommendations was tested based on color infrared aerial photographs for the four site-years. Predicted tiller density (derived from the relationship between relative NIR and relative tiller density) vs. measured tiller density was plotted. This plot was then divided into quadrants using the critical threshold of 540 tillers m-2 (Weisz et al., 2001) for making GS-25 N recommendations, and the percentage of the data in each quadrant was computed. Data in the upper right and lower left quadrants represent instances when N management decisions based on color infrared aerial photography would have been correct. Data in the upper left and lower right quadrants represent the percentage of incorrect N management recommendations that would have been made based color infrared aerial photography.
|
RESULTS AND DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
|
|
Correlations between the R band and tiller density also differed somewhat between the two film types (Tables 2 and 3). Both film types had a significant correlation between R digital counts and tiller density at TRS in 1998 and failed to show a relationship between TRS in 1999 and Whitehat. The major difference between the two film types was at the Winslow farm where the color film showed a significant correlation between R digital counts and tiller density but the color infrared film did not. Leaf pigments and cell-wall cellulose are transparent in the NIR wavelengths, making reflectance in this region highly sensitive to plant biomass (Gates et al., 1965; Guyot, 1990). This may be why NIR digital counts had the highest and most consistent correlations with tiller density (Table 3). The color film included a small portion of the NIR wavelengths in the R band (590–730 nm), and that may explain why R digital counts from the color film had slightly better sensitivity to tiller density compared with the R band of the color infrared film.
The correlations between tiller density and normalized NIR, NDVI, RVI, DVI, SAVI, OSAVI, and NIR + R bands were also examined at all four locations (Table 4). Significant correlations were found between each of these indices and tiller density at TRS in 1998, TRS in 1999, and Winslow. No significant correlations between any of these indices and tiller density were found at the Whitehat site.
|
Based on the strength and consistency of the statistically significant correlations between tiller density and NIR digital counts at all locations, NIR was chosen for linear regression analysis (Fig. 1). The coefficients of determination for NIR digital counts regressed against tiller density at each location ranged from 0.45 to 0.76 (Table 5). If the intercepts and slopes of the linear regressions were similar, this would indicate that a single equation might be used to predict tiller density across environments. If the intercept or slope changed across environments or years, then some form of ground calibration would be required at each location. Unfortunately, there were significant differences in the slopes of the regressions of NIR digital counts against tiller density (Fig. 1 and Table 5) across sites.
|
|
In order to remove differences in the relationship between tiller density and NIR digital counts across environments, a procedure analogous to that described by Blackmer et al. (1996) was used. Table 6 shows that this procedure was successful in removing the differences across environments. The coefficients of determination for relative NIR digital counts regressed against relative tiller density at each location ranged from 0.57 to 0.88 (Table 6). Because the differences across environments were removed by this procedure, the data for each site-year was combined. This resulted in a linear relationship (R2 = 0.77) between relative NIR digital counts and relative tiller density of the pooled data with a slope of 1.04 and an intercept of 0.07 (Fig. 2 and Table 6).
|
|
 | (7) |
While the linear regression between relative NIR digital counts and relative tiller density had a R2 of 0.77, the most important factor for making N application decisions can be simplified to being above or below a critical threshold (Weisz et al., 2001). Equation [7] can be used to represent the data shown in Fig. 2, in terms of N application decisions. Figure 3 shows predicted tiller densities vs. measured tiller densities from the four site-years used in this study. Consistent with Weisz et al. (2001), a critical density threshold of 540 tillers m-2 is shown in Fig. 3. Seven percent of the samples were predicted to be below the threshold when they actually had higher tiller densities (lower right quadrant), and N would have been recommended when it was not required. Eleven percent of the samples (upper left quadrant) were erroneously predicted to have densities above the threshold, and N would have not been recommended when it was needed. In 82% of the samples (upper right and lower left quadrants), the correct N recommendation would have been made using Eq. [7] at GS 25.
|
|
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Our second objective was to determine if minimal ground truthing could be used to develop a single relationship between a measure of reflectance and tiller density across locations. A two-step procedure using relative NIR digital counts and relative tiller density was tested. This resulted in a linear relationship (R2 = 0.77) that could be used across locations and years (Fig. 2). By measuring high and low tiller density locations and determining the NIR digital counts at those locations, a consultant might use Eq. [7] to determine tiller density across a field. Finally, Fig. 3 shows the accuracy of using color infrared aerial photographs and Eq. [7] to make GS-25 N recommendations for the data collected. This resulted in the correct GS-25 N recommendation 82% of the time (Fig. 3).
The critical tiller density threshold used in Fig. 3 for making GS-25 N recommendations was based on previous research in this production region (Weisz et al., 2001). Higher thresholds (1000 tillers m-1) have been reported for other areas in the Southeast (Scharf and Alley, 1993). We did not test the accuracy of Eq. [7] for predicting above or below this higher threshold because it is higher than the upper range of tiller densities found in this study (Fig. 1 and 3). However, Fig. 3 demonstrates the strength of this approach for making GS-25 N recommendations. While it does not eliminate the need to scout fields and count tillers altogether, it does reduce the amount of ground sampling. Instead of counting tillers at many locations across fields, sampling at only two (high and low tiller density) areas is needed.
The data in this research were from wheat fields where weed, insect, and disease pests and plant nutrition were well managed. Growth stage 25 for timely planted wheat usually occurs in this region in late January and early February. At that time, weed pressure is usually low and insect and disease pests have generally not yet developed. The fact that GS 25 occurs at a time when crop conditions are likely to be more uniformly favorable from field to field for remote sensing is encouraging, but it does not preclude the need for caution in selecting wheat fields for remote sensing that are well managed and generally free of factors such as pests or nutritional problems, which could potentially confound estimation of tiller density from relative NIR digital counts.
|
ACKNOWLEDGMENTS |
|---|
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  R. P. Sripada, D. C. Farrer, R. Weisz, R. W. Heiniger, and J. G. White Aerial Color Infrared Photography to Optimize In-Season Nitrogen Fertilizer Recommendations in Winter Wheat Agron. J., October 15, 2007; 99(6): 1424 - 1435. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. P. Sripada, R. W. Heiniger, J. G. White, and A. D. Meijer Aerial Color Infrared Photography for Determining Early In-Season Nitrogen Requirements in Corn Agron. J., June 5, 2006; 98(4): 968 - 977. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. P. Sripada, R. W. Heiniger, J. G. White, and R. Weisz Aerial Color Infrared Photography for Determining Late-Season Nitrogen Requirements in Corn Agron. J., September 19, 2005; 97(5): 1443 - 1451. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. B. Phillips, D. A. Keahey, J. G. Warren, and G. L. Mullins Estimating Winter Wheat Tiller Density Using Spectral Reflectance Sensors for Early-Spring, Variable-Rate Nitrogen Applications Agron. J., May 1, 2004; 96(3): 591 - 600. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Flowers, R. Weisz, and R. Heiniger Quantitative Approaches for Using Color Infrared Photography for Assessing In-Season Nitrogen Status in Winter Wheat Agron. J., September 1, 2003; 95(5): 1189 - 1200. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Flowers, R. Weisz, R. Heiniger, B. Tarleton, and A. Meijer Field Validation of a Remote Sensing Technique for Early Nitrogen Application Decisions in Wheat Agron. J., January 1, 2003; 95(1): 167 - 176. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 N. Aparicio, D. Villegas, J. L. Araus, J. Casadesus, and C. Royo Relationship between Growth Traits and Spectral Vegetation Indices in Durum Wheat Crop Sci., September 1, 2002; 42(5): 1547 - 1555. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| The SCI Journals | Crop Science | Vadose Zone Journal | |||
| Journal of Natural Resources and Life Sciences Education |
Soil Science Society of America Journal | ||||
| Journal of Plant Registrations | Journal of Environmental Quality |
The Plant Genome | |||
